國立臺灣大學資訊網路與多媒體研究所Huang, Kan-LiKan-LiHuangTung, Yi-ShinYi-ShinTungWu, Ja-LingJa-LingWuHsiao, Po-KangPo-KangHsiaoChen, Hsien-ShuoHsien-ShuoChen2006-11-142018-07-052006-11-142018-07-052002-08http://ntur.lib.ntu.edu.tw//handle/246246/200611150121446In this paper, we present a frame-based characteristics extraction tool, which can retrieve meaningful information from MPEG-compressed bitstreams. Taking advantage of the extracted information, we present a general architecture for MPEG video transcoding, which can transcode a given MPEG video bitstream into various MPEG video bitstreams with different characteristics. We also implement an MPEG-2-to-MPEG-4 video transcoder, which not only can change the bitstreams’ syntax and bitrate but also can encode each semantic video segment as a group of pictures (GOPs), by using the techniques of flame type conversion and video shot boundary detection. Moreover, the proposed transcoder can intelligently select motion mode for each macroblock of B-Vops in an MPEG-4 video to avoid some unpredictable artifacts due to mis-selection of motion mode. This representative applicationapplication/pdf1210710 bytesapplication/pdfzh-TWHBTInGaPkink phenomenonscattering parameterjournal articlehttp://ntur.lib.ntu.edu.tw/bitstream/246246/200611150121446/1/8106.pdf